nb-whisper-large-v0.8

<!-- This model card has been generated automatically according to the information Keras had access to. You should probably proofread and complete it, then remove this comment. -->

nb-whisper-large-v0.8

This model is a fine-tuned version of NbAiLab/nb-whisper-large-v3-RC4 on the NbAiLab/ncc_speech_styling_v2 dataset. It achieves the following results on the evaluation set:

• step: 49999
• validation_nst_loss: 0.4309
• train_loss: 0.4828
• validation_nst_wer: 2.2211
• validation_nst_cer: 0.6758
• validation_nst_exact_wer: 2.7655
• validation_nst_exact_cer: 0.7592
• validation_clean_stortinget_no_loss: 0.7845
• validation_clean_stortinget_no_wer: 8.8323
• validation_clean_stortinget_no_cer: 5.6753
• validation_clean_stortinget_no_exact_wer: 11.6973
• validation_clean_stortinget_no_exact_cer: 6.1161

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 7e-05
• lr_scheduler_type: linear
• per_device_train_batch_size: 8
• total_train_batch_size_per_node: 32
• total_train_batch_size: 1024
• total_optimization_steps: 50,000
• starting_optimization_step: None
• finishing_optimization_step: 50,000
• num_train_dataset_workers: 32
• num_hosts: 32
• total_num_training_examples: 51,200,000
• steps_per_epoch: 7254
• num_beams: None
• weight_decay: 0.01
• adam_beta1: 0.9
• adam_beta2: 0.98
• adam_epsilon: 1e-06
• dropout: True
• bpe_dropout_probability: 0.2
• activation_dropout_probability: 0.1

Training results

Framework versions

• Transformers 4.36.2
• Datasets 2.16.0
• Tokenizers 0.15.0